OpenAI Competitor Intelligence Agent

Team

• David & Jack — Intelligence Agent (design, prompt engineering, output)
• Tatsuro & Ez — Evaluation Agent (scoring framework, automated testing)

Project Structure

competitor-intel-agent/
├── README.md                          ← You are here
├── SUBMISSION.md                      ← Final markdown submission (all 10 sections)
├── report_output/
│   └── openai_intel_report.html       ← Sample HTML report output
├── knowledge_base/
│   ├── openai_context.md              ← OpenAI's products, strategy, priorities
│   ├── competitor_profiles.md         ← Tier 1-3 competitor baselines
│   └── industry_benchmarks.md         ← Model leaderboards, pricing, market data
├── prior_reports/
│   └── (weekly reports saved here after each run)
├── prompts/
│   ├── agent_system_prompt.md         ← The full agent prompt (Section 2)
│   └── prompt_log.md                  ← Log of all prompt iterations
├── scripts/
│   ├── run_agent.sh                   ← One-command agent runner
│   └── generate_report.py            ← Python script to format HTML output
└── eval/
    └── (Tatsuro & Ez's evaluation framework goes here)

Quick Start

Prerequisites

• Claude Code installed (npm install -g @anthropic-ai/claude-code)
• Anthropic API key set: export ANTHROPIC_API_KEY=your_key
• (Optional) Exa API key: export EXA_API_KEY=your_key

Run the Agent

# Option 1: One-command run
chmod +x scripts/run_agent.sh
./scripts/run_agent.sh

# Option 2: Run directly with Claude Code
claude-code --system-prompt prompts/agent_system_prompt.md \
  "Generate the weekly competitive intelligence report for OpenAI. \
   Today's date is $(date +%Y-%m-%d). \
   Read the knowledge base files in knowledge_base/ for context. \
   Output a structured report following the schema in the system prompt."

After Running

1. Review the generated report
2. Save a copy to prior_reports/week_of_YYYY-MM-DD.md
3. Update knowledge_base/competitor_profiles.md with any new findings
4. Log the prompt used in prompts/prompt_log.md

How to Edit & Contribute

For Jack (Data Sources)

• Review and update knowledge_base/competitor_profiles.md — add any competitors or sources I missed
• Review knowledge_base/openai_context.md — make sure OpenAI's product descriptions are accurate
• Add specific Twitter/X accounts to monitor in the agent prompt

For Tatsuro & Ez (Evaluation)

• Create your evaluation framework in the eval/ directory
• The agent's output schema is defined in SUBMISSION.md Section 5
• Suggested approach: 10 test scenarios with known-correct answers, score on signal detection accuracy, threat-level assignment, and source citation quality

For Everyone

• All prompt changes should be logged in prompts/prompt_log.md
• If you change the agent prompt, update both prompts/agent_system_prompt.md AND SUBMISSION.md Section 2

Submission Checklist

• All 10 sections present in SUBMISSION.md
• Agent prompt is complete and not summarized
• At least one real sample output included
• At least one concrete failure example in Section 10
• Technologies justified (not just listed)
• HTML report renders correctly
• Prompt log shows iteration history

Timeline

• Saturday night: First draft (David)
• Sunday morning: Jack reviews data sources & knowledge base
• Monday 12pm Taiwan / 6pm Berkeley: Team sync (optional)
• Tuesday 5pm Berkeley / Wednesday 8am Taiwan: Final submission